Author: Nobuki Imano, Daisuke Kawahara, Misato Kishi, Yuji Murakami 👨🔬
Affiliation: Department of Radiation Oncology, Graduate School of Biomedical and Health Sciences, Hiroshima University, Graduate School of Biomedical and Health Sciences, Hiroshima University 🌍
Purpose: This study aims to develop a comprehensive Multi-score by integrating Radiomics-score (Rad-score), Gene-score derived from gene expression levels, and tumor environment Rad-score (TE-Rad-score) based on specific regions of interest (ROI), aiming to improve survival prediction accuracy in early-stage non-small cell lung cancer (NSCLC) patients and enable biological interpretation of tumor behavior through the integration of imaging features from localized tumor regions and gene expression information.
Methods: CT images from 109 early-stage NSCLC patients were analyzed. ROIs included the GTV, expanded/contracted GTV, and surrounding ring-shaped areas. Radiomics features and gene expression levels underwent dimensionality reduction, and selected features were scored. TE-Rad-scores were created for each ROI to identify imaging features linked to the tumor microenvironment. Survival risk was stratified into high- and low-risk groups using individual scores and a combined Multi-score, evaluated via Kaplan-Meier (KM) curves of overall survival and C-index. Subsequently, enrichment analysis was conducted between these two groups to explore the biological functions present in the tumor environment derived from imaging features.
Results: In the validation cohort (32 cases out of 109), p-values from the KM curves and C-index values were as follows: Rad-score (0.067;0.62), Gene-score (0.018;0.83), and Multi-score (0.0025;0.85). Results from the enrichment analysis revealed that the Gene-score was particularly associated with pathways related to tumor cell proliferation and metastasis in NSCLC, including the Ca²⁺ signaling pathway and the retinoic acid signaling pathway. Among the TE-Rad-scores, the score derived from the GTV+5mm region, which includes the area surrounding the tumor, highlighted the Wnt/β-catenin pathway, suggesting its potential involvement in NSCLC tumor progression and therapy resistance.
Conclusion: Survival prediction accuracy improved significantly with the Multi-score, outperforming individual scores. Enrichment analysis of Rad-score and TE-Rad-score provided insights into the tumor microenvironment at both whole-tumor and localized levels, identifying distinct imaging features linked to these regions.